#APSA2016 Twitter Analysis

By Eric C. Vorst

visualization-7-2

The 2016 American Political Science Association annual meeting in Philadelphia was a great place to highlight new research, to learn from our peers, and to make new professional connections. It also provided an exciting opportunity to gain new insight into how networks of discussion evolve in real time over the duration of a major academic conference. Data mining software, content analysis, and social network visualization tools were used to observe how communities of discussion evolved as the conference unfolded, to identify the emergence of key themes, and to map the extent to which these themes reached different communities within the network.  Ultimately, this project helps to provide a unique insight into what political scientists talk about during a political science convention.

Methods: Network Visualization and Content Analysis

To gain this insight, I mined all tweets mentioning #APSA2016 at the conclusion of each day throughout the conference and used Gephi for Macintosh to create a daily network graph.  Gephi is a robust network visualization tool that creates graphical representations of networks using a wide range of user-specified algorithms.  The foundation of these graphs lies in their non-deterministic nature: data points represent relationships between network users rather than representing values, such as those which you would find in a traditional Cartesian graph.  The relationship-based type of analysis offered by Gephi is particularly fitting, considering social media is rooted in the power of relationships.

A quick primer on interpreting network visualization graphs is helpful in understanding how to best make sense of the data.  Network visualizations are built upon nodes and edges between nodes.  Each person in a network appears as a node (or circle), while each connection between that user and another user appears as an edge (or line).  For the Gephi visualization algorithm I used, each node has a repulsive force against other nodes, while each edge between nodes creates a gravitational attraction between those nodes.  The size of the node represents its degree, which is a measure of the number of total connections into and out of the node.  The size of a node’s label represents its eigencentrality, which is a measure of the node’s influence based upon the influence of the nodes to which it is connected.  Last, different colors represent different neighborhoods of discussion, which are identified through the use of a modularity algorithm.

visualization-1-1

nodes and edges

visualization-1-2

neighborhoods of discussion by color

 

While networks with a small number of participants sending few messages tend to be fairly simple and relatively uninspiring, visible patterns of network communities begin to develop as the number of participants and messages increase.  A major benefit of this type of analysis is that it allows us to identify how different types of messages propagate throughout a social network which, in turn, tells us a lot about their relative reach and impact.  Ultimately, social network visualization gives us the ability to observe the power of strong, interconnected, and influential relationships within a defined system.

Frequency Analysis

From a pure frequency standpoint, the most popular days for #APSA2016 were Thursday (9/1) and Friday (9/2), with Saturday (9/3) coming in a close third. When looking at the most popular hour for #APSA2016 tweets across all days, the highest number of tweets occurred during the 2pm hour, followed closely by the 1pm and 3pm hours.  I also noticed that the volume of tweets during the 5pm hour was approximately 1/3 lower than the volume of tweets during the 4pm and 6pm hours.  Initially, I supposed this might have been due to a higher number of attendees eating dinner.  However, my wife pointed out that this could have been due to happy hour.  This makes sense, because that would be a time when more people would be doing face-to-face networking.

graph-1-1

Another noteworthy finding was that #APSA2016 tweet volume did not seem to increase much until after 11:00 am each morning.  Interestingly, this roughly coincided with the end of the 10:00 am panel sessions.  Also, there was relatively scant tweet activity in the mornings, with fewer than 50 tweets being sent during the 7am hour.  In fact, it looks like the night owls outnumbered the early birds: roughly three times as many tweets were sent at 10am than at 7am, and nearly ten times as many tweets were sent at the midnight hour than at 7am.

graph-1-2

Daily #APSA2016 Network Analyses

I found some interesting trends in the daily #APSA2016 networks as the conference unfolded.  The @WomenAlsoKnow node was extremely influential at the start of the conference (8/31-9/1), exhibiting a high degree and high eigencentrality.  Additionally, @WomenAlsoKnow saw a fairly central location in the network while reaching out to a number of different neighborhoods of discussion.  It was also interesting to note that Christina Wolbrecht (@c_wolbrecht) served as a bridge between Nate Silver (@natesilver538) and the Monkey Cage (@monkeycageblog).  Dr. Wolbrecht’s prominence across the #APSA2016 network during the first day suggests her contributions to The Big Next Questions in Gender and Politics roundtable and the accompanying #WomenAlsoKnowStuff message had a significant influence on the first two days of the convention.

On Day Three (9/2), the #APSA2016 network went absolutely bananas over the Monkey Cage Blog, which was featured in the well-received roundtable How to Write about Your Research for the Monkey Cage.  The @monkeycageblog node saw a very high degree and eigencentrality, along with a central location in the network reaching out to a wide range of neighborhoods.  @WomenAlsoKnow remained prominent in the network on Day Three, likely due to the continued buzz surrounding their #WomenAlsoKnowStuff messaging, while @APSAtweets (the American Political Science Association’s official Twitter account) maintained a stable and central location.

visualization-3-1visualization-3-2visualization-3-3

Day Four (9/3) featured some interesting architectural elements in the #APSA2016 network, as no single individual account seemed to dominate the center of discussion.  I found this particular day to provide great examples of the differences between eigencentrality, degree, and network centrality.  Specifically, a substantively interesting juxtaposition emerged between degree and eigencentrality, suggesting that while the #APSA2016 network didn’t necessarily eschew overtly political discussion, it didn’t favor it either.

A strong eigenvalue neighborhood formed around Daniel Kreiss (@KreissDaniel), who was a member of the Day Four roundtable Past, Present, and Future of Digital Politics.  This neighborhood featured several nodes who, while neither possessing extremely high individual degrees nor a predominantly central location in the larger network, were connected to and interconnected with influential members of the #APSA2016 network at a relatively higher frequency.

It was also intriguing that two of the three nodes with the highest degree on Day Four also ranked relatively low in eigencentrality and network centrality.  Each of these two nodes featured tweets that were overtly political in nature, with one criticizing the Republican Party for a lack of intellectual curiosity and the other criticizing political scientists for demonstrating an overly liberal ideological balance.  Both of these nodes experienced a relatively high frequency of traffic, yet there were very few roads connecting them to the more active neighborhoods.  Rather put, while there were instances where the frequency of messages containing political content did spike, these messages were confined within their own fairly polarized neighborhoods of discussion.  Despite a rise in frequency, there was not a corresponding rise in influence.

Full Conference Analysis

Having examined each individual day of the #APSA2016 network, I wanted to get a big picture view of the conference as a whole.  While a traditional approach would involve looking at the big picture first, this project evolved in real time.  Thus, the big picture was not available until the conference had come to a close.

First, I used a simple Excel “find and replace” function to perform a raw content analysis of all tweets over the duration of conference.  The top ten words across all tweets included: panel, political, booth, paper, politics, book, philly, research, thanks, and data.  However, this basic content analysis excluded words including hashtags and ampersands, which led to somewhat misleading results as it underrepresented the popular themes expressed through tweets mentioning #WomenAlsoKnowStuff and @WomenAlsoKnow.  Further, while raw frequency counts are useful in a general sense, they paint a fairly broad contextual picture that doesn’t tell us much about the interactions, relationships, and neighborhoods that host these discussions.

Expanding an analysis beyond mere frequency of tweets allows for an examination of their true reach and impact across the network.  To perform this portion of the #APSA2016 network analysis, I first isolated all tweets with the word “women” in any part of the message, then highlighted them in yellow in the network map.  First impression?  Those are a lot of tweets!  It was also interesting to note that the majority of these tweets seemed to be in the middle of the network, with fewer occurring in the peripheral areas.  I interpreted this to mean that tweets with the word “women” in them tended to be shared amongst the more influential members within the network.

visualization-5-1visualization-5-2

Next, I isolated tweets with the words “politics” and “political”, and highlighted them in teal.  In contrast to tweets mentioning the word “women”, tweets with the words “politics” and “political” had a much broader reach across the entirety of the #APSA2016 network.

When highlighting tweets mentioning “women” and tweets mentioning “politics” and “political” in magenta, it was interesting see how the combination of these terms interacted with each other in the network.

Finally, I highlighted tweets with the words “panel”, “paper”, and “research”.  Since these terms represented the core of our activities at the conference, I wanted to see how these terms mapped out in the #APSA2016 network.  There were a couple of things that stood out to me after doing so.  First, and not surprisingly, there was a high frequency of tweets mentioning these words connected to the @APSAtweets and @WomenAlsoKnow nodes.  This suggests discussion attached to these nodes was primarily directed towards the academic aspect of the conference.  Conversely, there were very few tweets mentioning “panel”, “paper”, or “research” connected to the @MonkeyCageBlog.  This indicates that while the Monkey Cage may have been one of the influential members of the #APSA2016 network, it was due to their influence in other types of messaging themes.

The full network analysis video for the entire #APSA2016 conference can be seen here:

Final Thoughts

Social media analysis doesn’t just require us to ask new questions; it requires us to ask questions in new ways.  I believe this opens up a great deal of creative possibilities that, if pursued with sufficient academic rigor, have the potential to establish a valuable beachhead in the study of contemporary political communication.  I am honored to be part of a community of scientists from a variety of disciplines who are forging new ground in the pursuit of better understanding the role social media plays in shaping the way citizens engage in the political process.  I’m very optimistic and excited about what the near future holds in this emerging subfield of scientific inquiry.

I hope my work can provide some motivation for others like me, who may be working with limited resources in the form of funding and computing power.  There is a lot to be said for pursuing good questions, looking at things from a unique perspective, and making creative use of whatever tools one may have at his or her disposal.  In the end, sometimes it’s just plain fun to grab on to an inspired idea, jump in with both feet, and see where it takes you!

Eric C. Vorst is a PhD candidate in Political Science at the University of Missouri – St. Louis, where he is writing his dissertation on incivility in social media during the 2016 presidential election.  Eric earned his M.A. in political science from University of Missouri – St. Louis, his M.B.A. from Lindenwood University, and his B.A. in English Literature and Philosophy from Central Missouri State University.  His research interests include political communication and behavior, network analysis, and American political development. Eric lives in the St. Louis area with his wife and two children, aged 3 and 5. Eric will be presenting “Mountains or Molehills? Examining the Trump Effect on Twitter” at the 2017 Western Political Science Association conference in Vancouver, Canada.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s